Author:

Keywords:

Science & Technology, Life Sciences & Biomedicine, Health Care Sciences & Services, Pharmacology & Pharmacy, Linear mixed model, Missing at random, Missing completely at random, Non-future dependence, Pattern-mixture model, Selection model, Shared-parameter model, PATTERN-MIXTURE MODELS, REGRESSION-MODELS, SELECTION MODELS, NONRANDOM DROPOUT, CATEGORICAL-DATA, MISSING DATA, INFERENCE, STRATEGIES, LIKELIHOOD, SCORE, 0104 Statistics, 1117 Public Health and Health Services, Statistics & Probability, 3214 Pharmacology and pharmaceutical sciences, 4905 Statistics

Abstract:

Statistical analysis often extends beyond the data available. This is especially true when data are incompletely recorded because ad hoc as well as model-based approaches are rooted not only in the observed data and the mechanism governing missingness, but also in the unobserved given the observed data. Other instances of this phenomenon include but are not limited to censored time-to-event data, random effects models, and latent class approaches. One needs to be aware of: (1) changes in results and intuition relative to complete-data analysis; (2) the assumptions under which such approaches are valid; (3) the sensitivities implied by departures; and (4) in response to these, what sensitivity analysis avenues are available. This article provides a bird's-eye perspective on these. Some of the developments are illustrated using data from a clinical trial in onychomycosis. Copyright © 2009 Drug Information Association, Inc. All rights reserved.